Breast cancer cells condition lymphatic endothelial cells within pre-metastatic niches to promote metastasis

Transcription

1 Received 24 Dec 213 Accepted 16 Jul 214 Published 2 Sep 214 DOI: 1.138/ncomms5715 Breast cancer cells condition lymphatic endothelial cells within pre-metastatic niches to promote metastasis Esak Lee 1,2, Elana J. Fertig 3, Kideok Jin 3, Saraswati Sukumar 3, Niranjan B. Pandey 1 & Aleksander S. Popel 1,2,3 Breast cancer metastasis involves lymphatic dissemination in addition to hematogenous spreading. Although stromal lymphatic vessels (LVs) serve as initial metastatic routes, roles of organ-residing LVs are underinvestigated. Here we show that lymphatic endothelial cells (s), a component of LVs within pre-metastatic niches, are conditioned by triple-negative breast cancer (TNBC) cells to accelerate metastasis. s within the lungs and lymph nodes, conditioned by tumour-secreted factors, express CCL5 that is not expressed either in normal s or in cancer cells, and direct tumour dissemination into these tissues. Moreover, tumour-conditioned s promote angiogenesis in these organs, allowing tumour extravasation and colonization. Mechanistically, tumour cell-secreted IL6 causes Stat3 phosphorylation in s. This pstat3 induces HIF-1a and VEGF, and a pstat3-pc-jun-patf-2 ternary complex induces CCL5 expression in s. This study demonstrates anti-metastatic activities of multiple repurposed drugs, blocking a self-reinforcing paracrine loop between breast cancer cells and s. 1 Department of Biomedical Engineering, Johns Hopkins University School of Medicine, Baltimore, Maryland 2125, USA. 2 Department of Chemical and Biomolecular Engineering, School of Engineering, Johns Hopkins University, Baltimore, Maryland 21218, USA. 3 Department of Oncology, The Sidney Kimmel Comprehensive Cancer Center, Johns Hopkins University School of Medicine, Baltimore, Maryland 21231, USA. Correspondence and requests for materials should be addressed to A.S.P. ( apopel@jhu.edu). NATURE COMMUNICATIONS 5:4715 DOI: 1.138/ncomms & 214 Macmillan Publishers Limited. All rights reserved.

2 NATURE COMMUNICATIONS DOI: 1.138/ncomms5715 The lymphatic endothelium (LE), which comprises lymphatic endothelial cells (s), is a specialized endothelium and is distinct from the vascular endothelium. It lacks erythrocytes in the lumen and a well-defined basement membrane 1. Due to the leaky nature of the LE, lymphatic vessels (LVs) function as a reservoir for the lymph fluid consisting of proteins and cells that have leaked from the vascular system, and transport it back from the tissues to the circulatory system. In cancer, however, the prevailing view is that LVs are routes for cancer metastasis 2. Numerous studies have shown that tumour LVs serve as initial routes for metastasis. However, mechanisms of lymphogenous metastasis and, particularly, roles of organresiding LVs in metastasis are not well understood, despite the broad distribution of the LVs throughout the body. Gene expression in s is distinct from those in blood endothelial cells (BECs) 3,4, thus LV-mediated metastasis could be modulated by -derived factors. For example, it is known that stromal s attract tumour cells into the LVs by expressing CXCL12 and CCL21, chemokine ligands of CXCR4 and CCR7; CXCR4 and CCR7 are chemokine receptors expressed in several types of cancer cells 5,6. We asked what other -derived factors, including chemokines, angiogenesis factors or cytokines, play a role in breast cancer metastasis, since we have observed that secretion profiles of s are diverse and abundant, comparable to those of MDA-MB-231 (referred to below as MB231 for brevity) breast cancer cells in reverse western assays for 55 angiogenesis-related factors and 31 chemokines (Supplementary Fig. 1). We previously showed that treatment of animals with tumourconditioned media () prepared from triple-negative breast cancer (TNBC) cells accelerates lung and lymph node (LN) metastasis 7. We employed two different subtypes of TNBC cell lines: mesenchymal-like MDA-MB-231 and basal-like SUM149 (ref. 8). In that study, we observed that the lungs and LNs from -treated animals had 2 4 times elevation in organ-residing s, implying increased lymphangiogenesis, compared with serum-free media ()-treated animals. Strikingly, the treated group also showed 3 1 times more metastases in those organs within 4 weeks in the MDA-MB-231 model and 6 weeks in the SUM149 model, which is significantly faster than treated animals as well as current spontaneous metastasis models that take more than 7 1 weeks 9. This unexpected increase in metastasis led us to hypothesize that there are unknown signalling pathways among three partners: tumour-secreted factors (), organ-residing s and tumour cells. In this study, we investigate how -induced organ-residing s influence metastasis and propose novel mechanisms of metastasis as well as possible targets for therapeutic intervention for metastatic breast cancer. Here we employ a tumour-conditioned model, which involves -treated s in vitro or in vivo; this simulates the pro-metastatic effects of tumour-secreted factors in advanced breast cancer patients. In this report, we document for the first time that s within pre-metastatic organs are conditioned by tumour-secreted factors, and start to express CCL5 and vascular endothelial growth factor (VEGF), facilitating tumour cell recruitment, extravasation and colonization. We show that interleukin 6 (IL6) secreted by the tumour cells activates Stat3 pathways in s, resulting in lymphatic expression of CCL5 and VEGF. We propose central players for TNBC metastasis and test diverse repurposed drug agents to inhibit metastatic disease. Results Tumour-conditioned s express CCL5. Tumour-conditioned s (MB231-s) were prepared by growing normal s (n-s) in 3% (:EGM ¼ 3:7; EGM, endothelial growth media). We discovered that expression of CCL5 and CXCL7 was highly increased in MB231-s, compared with n-s (Fig. 1a). Since CXCL7 was also expressed in MB231 cells (Supplementary Fig. 2a), we focused on CCL5. CCL5 expression in MB231-s plateaued at day 2 (Fig. 1b), showing very high expression of CCL5 compared with n-s and MB231 cells (Fig. 1c). Another TNBC cell line, SUM149, and an oestrogen receptor-positive (ER þ ) breast cancer cell line, MCF7, were also tested: SUM149- promoted CCL5 expression in s, however, MCF7- did not (Fig. 1d). We next checked for -induced CCL5 expression in vivo, employing athymic nude mice (female, 5 weeks, NCI) to minimize the effect of T lymphocytes on CCL5 expression. CCL5 is also known as RANTES (regulated upon activation, normal T cell expressed and secreted), since T lymphocytes express and secrete it 1. We injected 5 ml of or prepared from MB231, SUM149 or MCF7 breast cancer cells subcutaneously as previously described 7,11. Mouse LVs (mlvs) in the LNs and lungs from the animals treated with MB231- or SUM149- expressed mouse CCL5 (mccl5), whereas the mlvs in animals treated with MCF7- or did not (Fig. 1e g). Brain tissues where LVs are absent did not show mccl5 expression on MB231- treatment (Fig. 1e,f). We assessed the concentration of mccl5 in -treated animals (Supplementary Fig. 3). We did not inoculate tumour cells in these animals so that we could measure mccl5 that was induced only by the (Supplementary Fig. 3a). treatment induced mccl5 in these animals; more than 45 pg ml 1 mccl5 was present in the mouse plasma (Supplementary Fig. 3b). -induced mccl5 was not associated with alpha smooth muscle actin (asma), a marker of myofibroblasts or pericytes (Supplementary Fig. 2b). Possible association of mccl5 with mouse CD45 (mcd45)-positive leukocytes, mf4/8-positive macrophages and miba-1-positive-activated macrophages was also examined (Supplementary Figs 4 6). Leukocytes were ubiquitously detected in the lungs in both - and treated animals (Supplementary Fig. 4a,b). Among those leukocytes, Iba-1- or F4/8-positive macrophages were detected in -treated lungs and LNs (Supplementary Figs 4 and 5). Importantly, -induced mccl5 was not colocalized with the leukocytes and macrophages but was associated with LYVE-1- positive s (Supplementary Fig. 6). CCL5 expressed by MB231-s drives metastasis. We observed that CM obtained from MB231-s promotes MB231 cell migration (Supplementary Fig. 7a). CCL5 can interact with CCR1/3/5 (ref. 12), so we blocked CCR1 by BX513, CCR3 by SB and CCR5 by maraviroc to determine which of these receptors induces MB231 cell migration. Only maraviroc blocked MB231 cell migration (Fig. 2a,b). We confirmed that both MB231 and MDA-MB-231-luc-D3H2LN express CCR5 (Fig. 2c), suggesting that -secreted CCL5 triggers chemotaxis of MB231 cells. The effect of the CCR5 inhibitor was compared with that of anti-ccr7-neutralizing antibodies in MB231 cell migration assays (Supplementary Fig. 7b,c), because CCL21, a chemokine ligand for CCR7, is known as another inducer of lymphatic metastasis 5. Maraviroc blocked MB231 cell migration induced by MB231--CM, whereas the anti-ccr7 antibody blocked n--cm-induced migration, demonstrating that the CCL5- CCR5 axis is essential for tumour cell migration towards tumourconditioned s rather than towards physiological s. We next pretreated animals with or daily for 2 weeks, followed by inoculation of MDA-MB-231-luc- D3H2LN breast cancer cells into the upper inguinal mammary fat 2 NATURE COMMUNICATIONS 5:4715 DOI: 1.138/ncomms & 214 Macmillan Publishers Limited. All rights reserved.

3 NATURE COMMUNICATIONS DOI: 1.138/ncomms5715 ARTICLE 15 % chemokine change (protein) CCL CCL21 CCL28 CCL28 CXCL16 Chemerin CXCL5 CCL26 CX3CL1 CXCL16 Chemerin CXCL5 CCL26 CX3CL1 CXCL1 CCL14 CCL1 CXCL8 IL16 CXCL1 CXCL1 CXCL11 CCL14 Lymphotactin CCL1 CCL2 CXCL8 CCL7 CCL22 IL16 Midkine CXCL1 CXCL9 CXCL11 CCL3/4 Lymphotactin CCL2 CCL7 CCL22 Midkine CCL15 CCL2 CCL19 CXCL9 CCL3/4 CCL15 CCL2 CCL19 MB231 & secreted 24 h secreted CXCL7 CCL18 CXCL4 CCL5 CXCL12 CCL17 CXCL17 hccl5 (arbitrary unit) MB231- n Day hccl5 (pg ml 1 ) h n- MB231 MB ±.2 pg per 1, MB231- hccl5 (arbitrary unit) MB231- MCF7- SUM149- MB231 MCF7 SUM149 n- Brain mlyve-1 mccl5 Lymph node Lung mccl5 (arbitrary unit) LN Lung MB231 NS Brain MB231 mlyve-1 mccl5 mlyve-1 mccl5 MCF7 mlyve-1 mccl5 SUM149 Figure 1 Tumour-conditioned s (MB231-s) express CCL5. (a) Reverse western assays with the human chemokine antibody arrays (R&D Systems) detected the relative level of 31 chemokines expressed in n-s or tumour-conditioned s (MB231-s). MB231-s were prepared by growing n-s in 3% media for 4 days. The media were replaced with 3 ml with 2% FBS. After 48 h, the supernatant was centrifuged and filtered. The resulting MB231--CM (MB231--CM) were analysed, comparing with n--cm. (b) ELISAs for human CCL5 (Quantikine ELISA, R&D System) performed on MB231- and n--cm. MB231--CM and n--cm were obtained at days, 1, 2, 3 and 4 of induction, and we showed accumulation of CCL5 plateauing at day 2 (n ¼ 4). (c) CCL5 concentration in each CM was determined at 48 h by CCL5 ELISAs. CCL5 expression in MB231-s was significantly higher than that in n-s (P ¼.23) or in MB231 (P ¼.38). MB231-s (1,) expressed 1.2±.2 pg hccl5 in 48 h (n ¼ 3). (d) was treated with obtained from MCF7, MB231 and SUM149 cells. MB231- and SUM149- induced CCL5 expression in s, compared with the secretion from n-s (Po.1); however, MCF7- were inactive (n ¼ 3). (e) (5 ml) prepared from MB231, SUM149 and MCF7 cells or were subcutaneously administered into nude mice (4 5 weeks, female, NCI) for 2 weeks. Excised organs (brains, Br-LNs, lungs) were fixed, frozen, sectioned and probed with anti-mouse LYVE-1 and anti-mouse CCL5 antibodies. LNs and lungs from MB231--treated animals showed mccl5 expression around mlvs. No mccl5 expression was seen in the LNs and lungs from the -treated groups, and the brains from either group. LVs are absent in the brains. Scale bar, 1 mm. (f) mccl5 pixel density was quantified by ImageJ (P ¼.48, P ¼.75, n ¼ 12). (g) MCF7- or SUM149- was injected into the animals for two weeks after which lungs were collected, fixed, sectioned and probed with antimouse LYVE-1 and anti-mouse CCL5 antibodies. SUM149- treatment induced mccl5 expression in lungs, while MCF7- treatment did not. pads and treatment with maraviroc (8 mg kg 1 per day, per os (p.o.)) or vehicle (Supplementary Fig. 8a). At week 5, 9 out of 1 mice in the -treated group had metastases, while only 2 out of 1 mice in the -treated group had them. In the maraviroctreated group, only four mice had metastases showing the antimetastatic effect of maraviroc (Fig. 2d). Primary tumour growth was not influenced by the treatment (Fig. 2e). Maraviroc treatment inhibited metastasis in the lungs and LNs, as shown by the reduced photon flux in the organs (Fig. 2f,g). The hearts, brains, spleens and livers did not show significant metastases (Supplementary Fig. 8b,c). Next, the effect of maraviroc was assessed in spontaneous metastasis models without pretreatment. We showed potent prevention of lung and LN metastasis by maraviroc treatment in these models as well (Supplementary Fig. 9). These results demonstrate that the CCL5- CCR5 axis is pivotal for lung and LN metastasis in -induced and spontaneous metastasis models and that it can be targeted to inhibit metastasis. MB231-s have abnormal expression of angiogenesis factors. We discovered that subcutaneous matrigel (5 ml per injection) mixed with s (2 1 6 ) induced moderate intra-gel angiogenesis in vivo (Supplementary Fig. 1a). We screened for NATURE COMMUNICATIONS 5:4715 DOI: 1.138/ncomms & 214 Macmillan Publishers Limited. All rights reserved.

4 NATURE COMMUNICATIONS DOI: 1.138/ncomms h MB231- (control) Maraviroc BX513 SB % migration NS CCR5 GAPDH Mw (kda) Control BX513 Maraviroc SB MDA-MB-231 MDA-MB-231 -luc-d3h2ln Number of mice with thoracic met (N=1) (N=1) + Mara (N=1) Week + Mara Luminescence x1 5.8 Radiance (p/sec/cm 2 /sr) Mammary tumor volume (mm 3 ) + Maraviroc Week NS Lung (x1 3 ) Lymph node (x1 2 ) Metastasis photon flux (N=1) (N=1) Mara (N=1) (N=35) (N=39) Mara (N=38) + Mara Lung Luminescence x Radiance (p/sec/cm 2 /sr) + Mara Lymph node Luminescence x Radiance (p/sec/cm 2 /sr) Figure 2 MB231-s promote metastasis through the CCL5-CCR5 axis. (a) MB231 cells were pre-labelled with Cell Tracker Green and their migration was assessed using the Oris cell migration kit. Labelled MB231 cells (5,) in complete media were added to each well of a 96-well plate containing stoppers to prevent the cells from settling in the centre region of the wells. Cells were allowed to adhere for 4 h, after which the stoppers were carefully removed. MB231--CM with or without inhibitors were added, and the cells that migrated to the centre of the well were quantified by measuring the fluorescence at 485/53 nm (n ¼ 4). Maraviroc, a CCR5 inhibitor, potently blocked MB231 cell migration in the presence of MB231--CM at 18 h. Scale bar, 5 mm. (b) Fluorescent signal from the migrated cells from a was measured at 485/53 nm and quantified (P ¼.13, n ¼ 4). (c) Human CCR5 levels in 3, MB231 and luc-mb231 cells were measured by western blotting. GAPDH was used as a loading control. (d) Athymic nude mice (4 5 weeks, female, NCI, n ¼ 1) were pretreated with or (5 ml) for 2 weeks before inoculation with luc-mb231 tumour cells and initiation of maraviroc (8 mg kg 1 per day, p.o.) or vehicle treatment. Five weeks later, the maraviroc-treated group showed B5% inhibition of metastasis, compared with vehicle-treated group. Red circles represent thoracic metastasis observed with the IVIS imager. (e) Tumour volume was measured using a caliper (n ¼ 1), and the volume was calculated using the formula: V ¼.52 (length) (width) 2.(f) Quantification of g, luciferase-mediated photon flux from the lungs (n ¼ 1) and the LNs (n ¼ 35 39) were obtained by using Living Image 3D Analysis (Xenogen; P ¼.8, P ¼.42). (g) Representative organ images under the IVIS imager. Data (b,e,f) are reported as mean±s.e.m. Original gel images of data (c) are presented in Supplementary Fig. 25. NS, nonsignificant. angiogenesis-related factors in -CM, using a reverse western array spotted with antibodies for 55 angiogenesis-related factors (Supplementary Fig. 1b). -secreted pro-angiogenic factors (angiogenin, endothelin, HB-EGF, IGFBP-2, MMP-9, PDGF-AA, PlGF), inflammatory factors (CD26, IL-1b, IL-8, CCL2) and antiangiogenic factors (angiopoietin-2, endostatin, pentraxin-3, serpin-e1, TIMP-1, IGFBP-3) into the CM (-CM; Supplementary Fig. 1c). Although -CM moderately induced EC proliferation, the rate of proliferation was far smaller than that in EGM-2, suggesting that -secreted anti- and pro-angiogenic factors are in balance for angiogenic homeostasis or that secreted pro-angiogenic factors are not sufficient to trigger angiogenesis (Supplementary Fig. 1d,e). We hypothesized that the angiogenic homeostasis in s can be perturbed by treatment. To address this question in vivo, matrigels mixed with s (-matrigel group) were implanted into animals followed by systemic subcutaneous administration of or for 2 weeks (Fig. 3). For controls, HUVECmatrigel and no cell groups were prepared. Strikingly, profound intra-gel angiogenesis was observed in the -treated matrigel group (Fig. 3a). HUVEC group or no cell group showed relatively less angiogenesis. Tail-vein injection of fluorescein isothiocyanate (FITC)-dextran (7 kda) visualized angiogenesis in the plugs (Fig. 3b,c). Infiltration of the host blood vessels (BVs) into the plugs was also observed (Fig. 3d). Immunostaining with anti-mcd31 (Fig. 3e,f) and anti-lectin antibodies (Supplementary Fig. 11) showed that the recruitment of the host BVs was increased by. Anti-hVEGFR3 staining was performed to detect human s previously included in the matrigel plugs (Fig. 3e,f). 4 NATURE COMMUNICATIONS 5:4715 DOI: 1.138/ncomms & 214 Macmillan Publishers Limited. All rights reserved.

5 NATURE COMMUNICATIONS DOI: 1.138/ncomms5715 ARTICLE treated HUVEC No cell treated HUVEC No cell FITC intensity 16, 12, 8, 4, NS NS HUVEC No cell HUVEC No cell 25, NS Margin Invading BVs hvegfr3 mcd31 Arbitratry unit 2, 15, 1, 5, FITC-dextran perfusion mcd31 hvegfr3 % Factor change Angiogenin Ang-2 CXCL16 CD26 HUVEC EGF Endoglin Endostatin Endothelin FGF-1 FGF-2 GM-CSF HB-EGF IGFBP-2 IGFBP-3 IL-1β IL-8 CCL-2 MMP-9 Pentraxin 3 PDGF-AA PDGF-BB Persephin PIGF Prolactin Serpin E1 TIMP-1 TIMP-4 TSP-1 upa VEGF VEGF-C Increased factors 6 Angiogenic factors 2 Inflammatory factors CXCL16 (inflammatory): 126% CD26 (inflammatory): 132% Endoglin (angiogenic): 1% EGF (angiogenic): 627% MMP-9 (angiogenic): 51% PDGF-AA (angiogenic): 57% PDGF-BB (angiogenic): 292% VEGF (angiogenic): 626% Decreased factors 1 Angiogenic factor 1 Inflammatory factor 4 antiangiogenic factors IL-1β (inflammatory): 79% Endostatin (anti-angio): 48% PTX-3 (anti-angio): 4% TIMP1 (anti-angio): 54% Ang-2 (anti-angio): 71% HB-EGF (angiogenic): 42% hvegf (pg ml 1 ).9 ±.15 pg VEGF per 1, MB231-1, MB231- MB231-HUVEC n- hlyve-1 hvegf mcd31 pvegfr2 (Y1175) Figure 3 MB231-s promote angiogenesis in vivo. (a) Matrigel plug assays with s/huvecs or without cells. Matrigel (5 ml) containing s or HUVECs (2 1 6 per gel) and heparin (1 U per gel) was injected subcutaneously on the ventral side of both flanks of nude mice. or (5 ml) were subcutaneously administered daily for 1 days, the mice were killed and the gel plugs were excised and analysed. Scale bar, 1 mm. (b) One hour before killing, FITC-dextran (7 kda) was injected through the tail vein to visualize BVs in the gel plugs. Plugs were homogenized and the intensity of FITC was measured and normalized to the volume of the plug (P ¼.3, n ¼ 4). (c) Representative FITC images of the gel plugs under the fluorescent microscope. Scale bar, 2 mm. (d) A representative image showing the margin of the plugs. Mouse BVs (mbvs) infiltrate the plug. Scale bar, 5 mm. (e) Gel plugs were fixed, frozen, sectioned and stained with anti-hvegfr3 (green) and anti-mcd31 (red) antibodies to detect hs and mbvs. Scale bar, 5 mm. (f) Quantification of e. (P ¼.37, n ¼ 12). (g) Reverse western assays with human angiogenesis antibody arrays detected the relative changes of 55 angiogenesis factors in s and HUVECs after tumour conditioning. Profiles of -derived factors (enhanced or downregulated) are described in the box (the right panel). did not induce any significant changes in HUVECs (n ¼ 2). (h) hvegf concentration (pg ml 1 ) in each CM was determined by ELISAs (P ¼.84, n ¼ 3). MB231-s (1,) secreted.9±.15 pg of VEGF. Immunostaining of -matrigel plugs revealed that hvegf 165 (green) expression was colocalized with hs (hlyve-1, red). Scale bar, 2 mm (right panel). (i) -matrigel from -treated animals showed phospho-vegfr2 (Y1175, red) around areas that were positive for mcd31 (green) and hlyve-1 (blue) signals. Scale bar, 1 mm. Data (b,f,g,h) are reported as mean±s.e.m. NS, nonsignificant. hlyve-1 Merged To understand these in vivo results, angiogenesis factors expressed in s/human umbilical vein endothelial cells (HUVECs) after treatment were assessed and compared with n-/huvec secretomes (Fig. 3g). -derived angiogenic factors that increased after treatment were Endoglin, EGF, MMP-9, PDGF-AA, PDGF-BB and VEGF. At the same time, four anti-angiogenic factors, including endostatin, pentraxin-3 (PTX-3), TIMP-1, and angiopoietin-2 were decreased (Fig. 3g). The factors secreted by HUVECs did not change after treatment. VEGF was dramatically increased in MB231--CM as seen by enzyme-linked immunosorbent assays (ELISAs) (Fig. 3h, left). Immunostaining of -treated -matrigel plugs also showed that hvegf 165 is colocalized with hlyve-1-positive human s (Fig. 3h, right). Phospho-VEGFR2 (Y1175) was detected around hs and mbvs, showing that the -secreted hvegf 165 could activate VEGFR2 signalling pathways (Fig. 3i). Although EGF was highly expressed in MB231-s (Fig. 3g), NATURE COMMUNICATIONS 5:4715 DOI: 1.138/ncomms & 214 Macmillan Publishers Limited. All rights reserved.

6 NATURE COMMUNICATIONS DOI: 1.138/ncomms5715 its angiogenic activity was not significant (Supplementary Fig. 12). MB231-s show angiogenic phenotypes. EC proliferation, migration, adhesion and tube formation were assessed in MB231- -CM (Fig. 4a,b). MB231--CM promoted HUVEC proliferation, migration and adhesion, compared with n--cm (Fig. 4a). Although robust HUVEC tube formation was observed in MB231--CM, tube formation was relatively poor in the same CM (Fig. 4b), suggesting that the MB231--CM primarily promotes angiogenesis rather than lymphangiogenesis, which is consistent with very low VEGF-C expression in MB231-s (Fig. 3g). We next generated growth factor-depleted (GF-dep-) by using anti-hvegf and anti-hegfneutralizing antibodies (Supplementary Fig. 13a). HUVEC adhesion assays confirmed that the immunodepletion was successful (Supplementary Fig. 13b). The immunodepletion was %EC proliferation %EC migration EGM-2 Serum-free media %EC adhesion n- cond. media MB231- cond. media MB231- cond. media 2 h HUVEC Br-LN mcd31 mcd31 mcd31 (LN) GF-dep- 15, 12, 9, 6, 3, mlyve-1 mvegf 164 Br-LN GF-dep- GF-dep- mcd31 FITC-Dextran mcd31 FITC-Dextran Lung ZO-1 ZO-1 GF-dep- GF-dep- - GF-dep-- + anti-vegf 165 ZO-1 ZO-1 ZO-1 Phalloidin Phalloidin Control GF-dep- GF-dep- - GF-dep-- + anti-vegf 165 Figure 4 GF-dep- promote LN angiogenesis and enhance lung vascular permeability. (a) In the proliferation assays, 2, HUVECs per well were plated in 96-well plates and allowed to adhere overnight. On the following day, the media were exchanged with -CM (cond. media), EGM or. Three days later, proliferating cells were quantified at 45 nm by using the WST-1 reagent (P ¼.39, n ¼ 6). In the migration assays, 18 ml of EGM-2 or or MB231--CM were added to the bottom chambers (CIM-plates), HUVECs (45, cells per well) were added to the top chamber. The bottom and top chambers were combined, loaded in the RTCA system and the cell index was measured continuously at 2 h (P ¼.11, n ¼ 2). In adhesion assays, HUVECs (25, cells per well) in 1 ml of EGM-2 or or MB231--CM were added in E-plates, after which the plate was loaded into the RTCA system. Cell indices at 3 h were analysed (P ¼.42 at 3 h, n ¼ 2). (b) HUVEC and tube formation (at 2 h) was induced by MB231--CM in matrigel matrix. Scale bar, 2 mm. (c) Human growth factor (hvegf 165 /hegf)-depleted (GF-dep-) or were subcutaneously administered daily for 1 days, the nude mice were killed and brachial LNs (Br-LNs) were excised and analysed with anti-mouse CD31 antibodies (green). Scale bar, 1 mm. (d) Quantification of c (P ¼.32, n ¼ 12). (e) Br-LNs from GF-dep--treated animals were probed with anti-mvegf 164 (green); mlv (red). Scale bar, 5 mm. (f) GF-dep--treated animals were perfused with FITC-dextran (7 kda) 1 h before termination. Collected lungs were stained with anti-mcd31 (red); dextran (green). Scale bar, 1, mm. (g) Anti-ZO-1 antibody staining (green) of HUVEC monolayers treated with (control), and GF-dep-. disrupted EC junctions while GF-dep- did not because of the absence of hvegf 165. Scale bar, 5 mm. (h) Anti-ZO-1 (green) antibody and anti-phalloidin (red) staining. GF-dep--conditioned -CM (GF-dep--) promoted disruption of EC junction. This was blocked by anti-vegf 165 antibody treatment. Scale bar, 5 mm. Data (a,d) are reported as mean±s.e.m. 6 NATURE COMMUNICATIONS 5:4715 DOI: 1.138/ncomms & 214 Macmillan Publishers Limited. All rights reserved.

7 NATURE COMMUNICATIONS DOI: 1.138/ncomms5715 ARTICLE performed because both the containing hvegf 165 and hegf and the -induced mvegf 164 in the mouse can promote angiogenesis in vivo, thus complicating interpretation of angiogenesis effects. The use of the GF-dep- clarifies that -induced angiogenesis in vivo is caused by host-derived mvegf 164 rather than by hvegf 165 or hegf already present in (Supplementary Fig. 14a). After treating animals with GF-dep- or, brachial LNs were probed with anti-mcd31 antibodies (Fig. 4c). LNs from GF-dep--treated mice showed enhanced angiogenesis (Fig. 4d). mvegf 164 was detected around mlvs in the GF-dep- -treated LNs (Fig. 4e), but not in -treated LNs. mvegf 164 was not found in the asma-positive area, but colocalized with mlvs (Supplementary Fig. 14b). To measure lung vascular permeability, FITC-dextran (7 kda) was intravenously injected after tumour conditioning: extravasation of dextran into the lungs was facilitated by GF-dep- treatment (Fig. 4f). In vitro, although disrupted the integrity of EC junctions of a HUVEC monolayer compared with -treated controls, GF-dep- did not cause junction disruption (Fig. 4g), consistent with hvegf 165 depletion above (Supplementary Fig. 13a). However, CM prepared from s treated with GF-dep- ( GF-dep--s ) caused EC junction disruption in vitro, and anti-hvegf 165 treatment normalized it (Fig. 4h). We confirmed that the junction disruption was not caused by EC apoptosis using cleaved-caspase 3 antibodies (Supplementary Fig. 13c). Anti-mVEGF 164 treatment inhibits lung and LN metastasis. Lungs from GF-dep-- or -treated animals were probed with anti-human VEGF165 and anti-mouse VEGF164 antibodies (Supplementary Fig. 14c). Anti-mouse VEGF164 antibodies have very limited cross-reactivity to human VEGF 165 (o.4% according to R&D systems for the anti-vegf 164 antibody AF- 493-NA). hvegf165 was not detected in either group, but mvegf164 was detected around the mlvs in GF-dep-treated lungs, demonstrating that lacking hvegf 165 (white) influences the mlvs to express mvegf164 (green; Supplementary Fig. 14c). GF-dep--treated animals were systemically administered anti-mvegf 164 or anti-hvegf 165 antibodies (5 mg kg 1, intraperitoneal (i.p.), every 4 days) during the GF-dep- induction phase. We discovered that antimvegf164 treatment normalized vascular permeability in GFdep--treated lungs, whereas anti-hvegf 165 did not (Fig. 5a). Next, the anti-mvegf 164 antibody was tested in GF-dep-induced metastasis models like the one discussed above induced by complete (Fig. 2). Five weeks after tumour inoculation in the induced mice, lungs and LNs were collected to assess metastases ex vivo (Fig. 5b). Anti-VEGF 164 antibody inhibited metastasis in the LNs and lungs (Fig. 5b,c), demonstrating that lung vascular remodelling and LN angiogenesis are initiated by GF-dep--induced VEGF, and the blockade of the VEGF function prevents metastatic extravasation and colonization. Dual inhibition of CCR5/VEGF strongly blocks metastasis. We established MB231 tumour xenografts (n ¼ 1) without pretreatment, and collected plasma at 2, 3, 4 and 5 weeks to estimate human tumour xenograft-induced mouse VEGF and mccl5 expression (Fig. 5d). Plasma samples from normal mice without tumours (n ¼ 8) were used as controls. Plasma concentration of mccl5 and mvegf was increased as tumours grew, compared with normal mice: mccl5 plasma concentration was 259.2±43.6 pg ml 1, and mvegf was 56.1±4.9 pg ml 1 when the mean tumour volume was 1,232±223 mm 3 (week 5). We hypothesized that dual inhibition of CCR5 and VEGF signalling would inhibit metastasis more effectively than single inhibition of each target, as the mccl5 and mvegf function as tumour recruitment factor and colonization factor, respectively (Fig. 5e). We carried out dual inhibition of CCR5 and VEGF as described in Supplementary Fig. 15. We observed that 6% of the mice had metastases in the anti-mvegf 164 group, 4% had metastases in the maraviroc group and only 2% had metastases in the combination group. All the mice (1%) had thoracic metastasis in the no-treatment group (Fig. 5f). The IL6-Stat3 axis induces CCL5 expression in s. We next identified key targets in tumour-conditioned s, which are specifically phosphorylated by treatment. Among 46 kinase phosphorylation sites screened, both S727 and Y75 of Stat3 were exclusively phosphorylated in s by treatment (Fig. 6a). The presence of phospho-stat3 (pstat3: Y75) in -treated s was confirmed in separate western blots (Fig. 6b). Importantly, the essential role of pstat3 in CCL5 expression in s was confirmed by using a small molecule, Stattic, an inhibitor of phosphorylation and dimerization of Stat3 (ref. 13) (Fig. 6c,d). We next showed that IL6 and granulocyte macrophage colonystimulating factor (GM-CSF) are exclusively expressed in TNBC cell lines (MB231 and SUM149), but not in MCF7 or s (Fig. 6e,f). GM-CSF was not considered as a key cytokine in the metastatic process because GM-CSF is known to phosphorylate Stat5 14 and we saw no pstat5 in -treated s (Fig. 6a). Human IL6 in MB231/SUM149/MCF7- and -CM was measured by ELISAs. High levels of IL6 were only seen in the TNBC cell lines (Fig. 6g). Only containing IL6-induced pstat3 in s and IL6-dep- ( immunodepleted of IL6) failed to induce phosphorylation of Stat3 in s (Fig. 6h; Supplementary Fig. 16a,b,d). These data demonstrate that TNBC cell-secreted IL6 is the crucial factor for induction of Stat3 phosphorylation in s. We also showed that the IL6-gp13- Jak2-Stat3 axis is critical for IL6 signal transduction (Supplementary Fig. 17). In functional assays, IL6-dep- did not induce CCL5 expression in s compared with intact, but still induced some VEGF expression (Fig. 6i). These results show that CCL5 expression in MB231-s is totally IL6 driven, but VEGF expression can be induced by IL6 and other unknown factors in the. To establish the relevance of these results to human disease, we analysed The Cancer Genome Atlas (TCGA) mrna-sequencing data from TNBC and estrogen receptor þ /progesterone receptor þ /human epidermal growth factor receptor 2 (ER þ /PR þ /HER2 ) tumours and discovered higher levels of expression of IL6 and CCL5 in TNBC (Supplementary Fig. 18a,b). Moreover, IL6 and CCL5 are significantly associated with LNpositive breast cancer in TNBC (Supplementary Fig. 18c,d), suggesting that the IL6-CCL5 axis that we discovered has clinical relevance for metastatic breast cancer patients. pstat3-pc-jun-patf-2 complex and HIF1a express CCL5/ VEGF. We observed that Stattic inhibited IL6-induced CCL5 and VEGF expression in s; SP6125, a c-jun N-terminal kinase inhibitor, blocked IL6-induced expression of CCL5 but not of VEGF in s (Fig. 7a). Western blots showed that c-jun and ATF-2 were constitutively phosphorylated in s, while Stat3 phosphorylation required IL6 (Fig. 7b). SP6125 reduced the amount of pc-jun and patf-2 but pstat3 was not affected. With Stattic, pstat3 disappeared but pc-jun and patf-2 were maintained (Fig. 7b). We next performed co-immunoprecipitation with nuclear extracts. Strikingly, pstat3, pc-jun and patf-2 form a ternary complex in response to IL6 treatment (Fig. 7c). NATURE COMMUNICATIONS 5:4715 DOI: 1.138/ncomms & 214 Macmillan Publishers Limited. All rights reserved.

8 NATURE COMMUNICATIONS DOI: 1.138/ncomms5715 epithelial cells 15. ATF-2 binds to the CRE 16. Moreover, c-jun and ATF-2 have been observed in a binary complex 17. Importantly, Stat3 can interact with c-jun and participate in cooperative transcriptional activation 18. We hypothesized that the pstat3-pcmcd31 FITC-Dextran Lung GF-dep- GF-dep- + anti-vegf 164 GF-dep- + anti-vegf 165 Lung GF-dep- Anti- VEGF 164 Luminescence 15 1 Lung ( 1 3 ) LN ( 1 2 ) LN 5 (N=1) GF-dep- (N=1) Anti-VEGF164 (N=1) (N=36) GF-dep- (N=38) Anti-VEGF164 (N=39) Counts 35 Normal mice (N=8) (below detection limit) Metastatic tumours Lymph nodes (LN) & lungs mccl5 (pg ml 1 ) Tumour bearing mice (N=1) Tumour size (mm 3 ) Tumour conditioning CCL5 (+) α-vegf Deregulated angiogenic factors Maraviroc Cancer cell migration Lung vascular permeability LN angiogenesis Recruitment Extravasation LN & lung metastasis Niche formation mvegf (pg ml 1 ) Normal mice (N=8) 11.6 ± 2.1 pg ml 1 Tumour bearing mice (N=1) Number of mice with thoracic met GF-dep- Maraviroc Anti-VEGF Tumour size (mm 3 ) Week Combination Figure 5 Anti-mVEGF 164 and maraviroc treatment inhibits LN and lung metastasis. (a) Nude mice were treated with GF-dep- or for 2 weeks. In addition, anti-hvegf 165 or and anti-mvegf 164 antibodies (i.p. injection, 5 mg kg 1, at days 1, 5, 1, 14) were administered. FITC-dextran perfusion showed that lung vascular permeability was increased by GF-dep- treatment. The increased permeability was normalized not by anti-hvegf 165 but by anti-mvegf 164. Scale bar, 1, mm. (b) Based on the result in a, we administered anti-mvegf 164 antibodies to inhibit GF-dep--induced metastasis. Nude mice were treated with or GF-dep- or GF-dep- þ Anti-mVEGF 164 antibodies for 2 weeks (i.p. injection, 5 mg kg 1, at days 1, 5, 1, 14). After 2 weeks, luc-mb231 tumour cells were orthotopically inoculated in the inguinal mammary fat pads. After 5 weeks of tumour inoculation, LNs and lungs were excised, incubated in D-luciferin solution and imaged under the IVIS imager. (c) Averaged photon flux in the lungs (n ¼ 1) and LNs (n ¼ 36 39) was quantified by using Living Image 3D Analysis (Xenogen; P ¼.33, P ¼.6). (d) Plasma concentration of mccl5 and mvegf in mice with and without MB231 tumours (n ¼ 8 1). Mouse plasma was obtained at 2, 3, 4 and 5 weeks after tumour inoculation. At week 5, when the tumour size was around 1,2 mm 3, the plasma concentration of mccl5 and mvegf was B26 and 56 pg ml 1, respectively (n ¼ 3). (e) Conceptual figure of tumourconditioned s mediated LN and lung metastasis. Tumour-conditioned s express CCL5, which induces tumour cell recruitment, and VEGF, which promotes angiogenesis and tumour extravasation. Blocking each target inhibits LN and lung metastasis. (f) Dual inhibition of VEGF and the CCL5-CCR5 axis. There are five groups described: not conditioned ( treated), tumour conditioned (GF-dep-), maraviroc treated, anti-mvegf 164 treated and combination group. We treated with anti-mvegf 164 antibodies (i.p. injection, 5 mg kg 1 per 4 days) during GF-dep- induction; maraviroc (8 mg kg 1 per day, p.o.) after two weeks of GF-dep- induction until the end of the experiment (n ¼ 1). Data (c,d) are reported as mean±s.e.m. After treating with SP6125 or Stattic, the complexes disappeared (Fig. 7c). The camp-responsive element (CRE) in the promoter of the CCL5 gene is known to regulate its expression in alveolar 8 NATURE COMMUNICATIONS 5:4715 DOI: 1.138/ncomms & 214 Macmillan Publishers Limited. All rights reserved.

9 NATURE COMMUNICATIONS DOI: 1.138/ncomms5715 ARTICLE Pixel treated EGM-treated -treated pstat3 STAT3 EGM Mw (kda) Control p38α(t18/y182) ERK1/2(T22) JNK pan GSK-3 (S21/S9) EGFR(Y168) MSK1/2(S376) AMPKα1(T174) Akt(S473) TOR(S2448) CREB(S133) HSP27(S78/S82) AMPKα2(T172) β-catenin Src(Y419) Lyn(Y397) Lck(Y394) STAT2(Y689) STAT5α(Y694) Fyn(Y42) Yes(Y426) Fgr(Y412) STAT6(Y641) STAT5b(Y699) Hck(Y411) Chk-2(T68) FAK(Y397) PDGFRβ(Y751) STAT5a/b(Y694) PRAS4(T246) p53(s392) Akt(T38) p53(s46) p7 (T389) p53(s15) c-jun(s63) p7 S6 kinase RSK1/2/3(S38) enos(s1177) STAT3(S727) p27(t198) PLC-γ1(Y783) STAT3(Y75) WNK1(T6) HSP6 PYK2(Y42) pstat3 STAT3 GAPDH MB231 EGM + DMSO DMSO Stattic Stattic (2μM) (5μM) SUM149 G-CSF IL1α IL1ra IL6 GM-CSF sicam1 MIF GROα IL8 Serpin E1 MCP-1 C5/C5a MW (kda) 37 MCF7 Relative CCL5 level EGM.1 Stattic (5μM) Time (hour) hil6 (pg ml 1 ) MB231 SUM149 MCF7 MCF7 SUM149 MB231 C5/C5a GROα MCP-1 MIF GM-CSF GROα MIF G-CSF sicam-1 IL8 Ctrl SerpinE1 IL6 IL8 Ctrl SerpinE1 GM-CSF GROα IL-1α IL-1ra IL6 IL8 Ctrl MIF sicam-1 Ctrl MIF Anti-IL6 pstat3 STAT3 GAPDH Mw (kda) 37 hccl5 (arbitrary unit) 48 h hvegf (arbitrary unit) 48 h EGM- - IL6-dep--.. Figure 6 Tumour cell-secreted IL6 phosphorylates Stat3, which induces CCL5 and VEGF expression in s. (a) Reverse western assays with the human phospho-kinase antibody arrays (R&D systems) simultaneously detected the relative amounts of 46 phosphorylation sites in s. We compared the effects of, EGM and treatment in s (overnight incubation). (b) Phosphorylation of Stat3 in -treated s was assessed in a separate western blot. (c) Phosphorylation of Stat3 was completely blocked by Stattic (5, 2 mm) in s. GAPDH was used as a loading control. (d) CCL5 levels were assessed by ELISA following 5 mm Stattic treatment (Po.1, P ¼.8, n ¼ 4). (e) Reverse western assays with the human cytokine antibody arrays (R&D systems) detected the relative amounts of 36 cytokines in MB231, SUM149, MCF7 and -CM. Representative images of the cytokine array membranes. (f) Summary of the cytokine array results in e. IL6 and GM-CSF were expressed only in TNBC cell lines (MB231 and SUM149). (g) ELISA was used to determine levels of IL6 in MB231 and SUM149 cells (n ¼ 3). (h) Stat3 phosphorylation was assessed in the presence or absence of anti-il6 in s. GAPDH was used as a loading control. (i) CCL5 and VEGF expression was measured by ELISA in the presence or absence of anti-il6 in s. VEGF expression was significantly reduced by IL6 depletion, but not completely (P ¼.75, P ¼.32, n ¼ 3). Data (a,d,g,i) are reported as mean±s.e.m. Original gel images of data (b,c,h) are presented in Supplementary Fig. 25. Jun-pATF-2 ternary complex would bind to the CRE site in CCL5 promoter. To test this hypothesis, we performed chromatin immunoprecipitation (ChIP) assay with B2-base pair chromatin fragments by sonication of s treated with IL6 (1 ng ml 1 ; Fig. 7d; Supplementary Fig. 19). Three regions of the CCL5 promoter with the CRE site ( 316 to 69 bp) and two distal sites ( 1,64 to 815 and 474 to 711 bp) were tested (Supplementary Fig. 19a). We found that pstat3-pc-junpatf 2 ternary complex specifically bound to only the CRE site (site 2) by real-time quantitative PCR. In contrast, the distal sites (sites 1 and 3) do not show significant complex-binding capabilities (Supplementary Fig. 19b,c). Compared with vehicle treatment, ChIPs on s with IL6 treatment showed specific pstat3-pc-jun-patf-2 ternary complex enrichment for binding to this region (Fig. 7d). In addition, electrophoretic mobility shift assays (EMSAs) were performed to show binding between the ternary complex and the CRE oligonucleotide (Fig. 7e). When s were treated with IL6, nuclear proteins bound to the CRE oligonucleotide; however, Stattic or SP6125 treatment inhibited the binding. No binding was observed on the mutated CRE, and excess unlabelled CRE oligonucleotide competitively inhibited the binding (Fig. 7e). VEGF expression in -treated s can be triggered by multiple signalling pathways, since IL6 depletion did not completely inhibit VEGF expression (Fig. 6i). However, we observed that IL6 promoted the expression of HIF-1a in NATURE COMMUNICATIONS 5:4715 DOI: 1.138/ncomms & 214 Macmillan Publishers Limited. All rights reserved.

10 NATURE COMMUNICATIONS DOI: 1.138/ncomms5715 % Protein expression Relative fold enrichment 12 1 %hccl5 %hvegf NS Targeting of IL6 and pstat3 blocks LN and lung metastasis. The mechanistic results above indicate that the IL6-Stat3 axis is a key inducer of CCL5 and VEGF expression in s. Thus, we tried to inhibit GF-dep--induced metastasis by targeting IL6 and pstat3 as described in Supplementary Fig. 2b. We generated GF/IL6-dep- by immunodepleting IL6 from GF-dep-. Separately, we chose S3I-21, another pstat3 inhibitor with the same mechanism of action as Stattic; S3I-21 has been tested in vivo 19. We showed that S3I-21 inhibited pstat3 levels in IL6- EGM DMSO Stattic SP IL6 (1 ng ml 1 ) EGM pc-jun Relative fold enrichment pstat3 pc-jun patf-2 GAPDH patf-2 Relative fold enrichment IL6 (1 ng ml 1 ) EGM DMSO SP Stattic Mw (kda) IL6 EGM IL6 EGM IL6 Probe: CRE in RANTES promoter WT(5 3): MUT(5 3): pstat IL6 (1 ng ml 1) Stattic (μm) SP6125 (μm) IP: patf-2 IP: pc-jun IP: pstat3 pstat3 HIF1-α GAPDH EGM WB: pstat3 WB: pc-jun WB: pstat3 WB: patf-2 WB: patf-2 WB: pc-jun Lamin B Mw 4 (kda) IL6 (1 ng ml 1 ) EGM DMSO SP Stattic Mw (kda) CRE (WT) CRE (MUT) CRE (WT) VEGF No VEGF/CCL5 IL6 (1 ng ml 1 ) Stattic SP IL6 (1 ng ml 1 ) Cold probe (times) EGM Stat P Jun P ATF No VEGF/CCL5 IL6 HIF P Stat P Jun P ATF CCL5 P Stat Stattic SP6125 Stat P Jun P ATF P Stat HIF Jun ATF VEGF Figure 7 The pstat3-pc-jun-patf-2 ternary complex is central for CCL5 expression and pstat3-dependent HIF-1a induces VEGF expression. (a) CCL5/VEGF expression in IL6-treated. were treated with EGM, IL6 only or IL6 with inhibitors (Stattic, stat3 inhibitor; SP6125, JNK inhibitor). IL6-dependent CCL5/VEGF expression was assessed using ELISAs. CCL5 with Stattic (P ¼.9, n ¼ 3); CCL5 with SP6125 (P ¼.28, n ¼ 3), VEGF with Stattic (P ¼.41, n ¼ 3). (b) Western blot assays with, EGM activated pc-jun and patf-2 but not pstat3; only IL6 treatment induced pstat3. SP6125 blocked pc-jun and patf-2 while Stattic selectively blocked pstat3. (c) Co-immunoprecipitation assays with nuclear extracts. Lamin B1 was used as a nuclear extract loading control. (d) ChIP assays, real-time PCR analysis of recruitment of patf-2, pc-jun and pstat3 to the CRE region (site 2) of CCL5 promoter with IL6 (1 ng ml 1 ) treatment (Po.5, n ¼ 3). (e) The two strands of the wild-type CRE oligonucleotide and of mutated CRE conjugated with biotin were synthesized (Invitrogen). For the binding reaction, 3 mg nuclear extract and.5 mg poly(di-dc) with or without excess unlabelled CRE oligonucleotide were incubated in binding buffer for 1 min at RT, after which oligonucleotide-biotin was added and incubated for 3 min at RT. Ten microlitres of binding sample was mixed with TBE sample buffer (Invitrogen), loaded on the gel and run for 1 h at 12 V in.5 TBE running buffer. The gel was transferred to a DNA transfer stack (Invitrogen). The nylon membrane was dried and cross-linked under a UV source (35 nm) for 15 min, then probed by the Chemiluminescent Nucleic Acid Detection Module (Pierce). (f) HIF-1a and pstat3 levels were assessed by western blot in the presence of IL6 treatment. (g) Graphical summary. pc-jun-patf-2 binary complexes and unphosphorylated Stat3 are present in n-s but there is no CCL5/VEGF expression. IL6 induces Stat3 phosphorylation and activates formation of the pstat3-pc-jun-patf-2 ternary complex, which is essential for CCL5 expression. pstat3 promotes HIF-1a expression and separately induces VEGF expression. On Stattic treatment, pstat3 and the ternary complex disappear, resulting in no expression of CCL5 and VEGF; the pc-jun-patf-2 binary complex that remains on Stattic treatment does not induce either CCL5 or VEGF expression. SP6125 dissociates both ternary and binary complexes, but pstat3 separately induces HIF-1a and VEGF expression. Data (a,d) are reported as mean±s.e.m. Original gel images of data (b,c,f) are presented in Supplementary Fig. 25. NS, nonsignificant. s (Fig. 7f), and this expression was blocked by Stattic but not by SP6125, demonstrating that HIF-1a expression is pstat3 dependent, but not associated with pc-jun or patf-2. Summarizing these immunoblot results (Fig. 7b f) and VEGF/ CCL5 expression data (Fig. 7a), we can conclude that the IL6-induced pstat3-pc-jun-patf-2 ternary complex is essential for CCL5 expression; VEGF expression is pstat3 induced and possibly HIF-1a associated and does not require pc-jun or patf-2 (Fig. 7g). 1 NATURE COMMUNICATIONS 5:4715 DOI: 1.138/ncomms & 214 Macmillan Publishers Limited. All rights reserved.

11 NATURE COMMUNICATIONS DOI: 1.138/ncomms5715 ARTICLE Number of mice with metastases (N=9) GF-dep- GF/IL6-dep- S3I Week GF/IL6-dep- GF/IL6- dep- Lung ( 1 3 ) LN ( 1 3 ) 1, 1, 1 1 1, 1, 1 1 N=9 N=28 32 Luminescence 12, 1, 8, 6, 4, 2, GF-dep- S3I-21 Lung LN (cortex) LN (medulla) Counts GF-dep- Lung LN (cortex) LN (medulla) S3I-21 Lymph node (LN) Breast tumour LN angiogenesis Lung Lung vascular permeability Tested in vivo Tested in vitro Anti-IL6 antibody IL6 IL6Rα VEGF Breast cancer cell P Stat3 gp13 gp13 Anti-VEGF antibody CCR5 P Stat3 Stat3 P Stattic/S3I-21 Maraviroc Cancer cell recruitment CCL5 P Stat3 Stat3 P P Jun P ATF CRE HIF1a Nucleus Organresiding Figure 8 Inhibition of IL6 and pstat3 blocks GF-dep--induced LN and lung metastasis. (a) There are four groups described: treated, tumour conditioned (GF-dep-), tumour conditioned without IL6 and pstat3 inhibited (by using a pstat3 inhibitor, S3I-21 ). After 2 weeks of these treatments, inguinal primary tumour was established, and thoracic metastasis was monitored for 5 weeks (as described in Supplementary Fig. 2b). The number of mice in each group with thoracic metastasis is shown. Lung metastasis was blocked by both IL6 depletion and S3I 21 treatment (n ¼ 9, Po.1). LN metastasis was blocked by IL6 depletion and by S3I 21 treatment (n ¼ 28 32, P ¼.27, P ¼.7). (b) Representative images of live animals in the IVIS imager, and images of collected lungs. Black arrows represent tumour nodules. (c) Immunohistochemistry with anti-cytokeratin antibodies on the lungs and LNs (cortex and medulla). Metastatic colonies are delineated with red-dotted curves. Scale bar, 5 mm. (d) Graphical summary of the whole study. Breast cancer cells secrete IL6 that interacts with IL6 receptor on s within the lungs and LNs. Activated IL6 receptors transduce the signals through gp13, phosphorylating lymphatic Stat3. pstat3 translocates into the nucleus to form the pstat3-pc-jun-patf-2 ternary complex, which is essential for CCL5 expression by targeting the CRE region in its promoter. pstat3 also induces HIF-1a to ultimately express VEGF in s. -secreted CCL5 recruits CCR5-positive breast cancer cells into the lymphatic system. The secreted VEGF enhances lung vascular permeability and induces LN angiogenesis to promote metastatic extravasation and colonization. The four possible targets in the overall mechanism were blocked as follows: by anti-il6 antibody (to target IL6), by maraviroc (to interrupt the CCL5-CCR5 axis), by anti-vegf antibody (to block LN angiogenesis and lung vascular leakiness) and by Stattic/S3I-21 (to inhibit pstat3). Treatments are marked with red and green circles, which represent tested in vivo and tested in vitro respectively. NATURE COMMUNICATIONS 5:4715 DOI: 1.138/ncomms & 214 Macmillan Publishers Limited. All rights reserved.

12 NATURE COMMUNICATIONS DOI: 1.138/ncomms5715 treated s (Supplementary Fig. 2a). All mice pretreated with GF-dep- for 2 weeks before tumour inoculation developed metastases at 5 weeks; 44% of the mice (4/9) treated with S3I-21 during the pretreatment phase developed metastases, only 22% (2/9) of the mice pretreated with GF/IL6-dep- had metastases that was less than the 33% (3/9) of mice with metastases in the -treated group (Fig. 8a). We observed significant reductions in lung and LN metastases by IVIS imaging, macroscopic morphology and anti-cytokeratin immunostaining (Fig. 8b,c; Supplementary Fig. 2c). Tumour size was not influenced by these treatments (Supplementary Fig. 2d). The mechanisms presented in this study are summarized in the schematic (Fig. 8d). Discussion According to the seed and soil hypothesis, metastatic cancer cells function as seeds and a particular organ microenvironment serves as the soil 2. It is difficult for cancer cells ( seeds ) to survive outside their site of origin, thus they have to find a suitable location ( soil ) where they can settle and grow. They also manipulate the microenvironment to optimize these premetastatic locations 21. In this study, we show for the first time that tumour cell-secreted IL6 conditions s in the pre-metastatic organs to prime them and promotes breast cancer metastasis. Paracrine signals regulated by the IL6-Stat3 axis and operating between cancer cells and s play a crucial role in the induction of CCL5 and VEGF expression in s within pre-metastatic organs facilitating tumour cell recruitment, extravasation and colonization (Fig. 8d). IL6 is an inflammatory cytokine that leads to activation of the Jak family and glycoprotein 13 (gp13) to phosphorylate Stat3 on interaction with the IL6 receptors 22. In our experiment using s, we showed that gp13, Jak2 and Stat3 were phosphorylated by containing IL6 (Supplementary Fig. 17). Stat3 is a transcription factor that contributes to the expression of diverse cytokines, chemokines and growth factors 23,24. Thus, the IL6- Stat3 axis has been explored in cancer The IL6-Stat3 axis promotes tumorigenesis 28 31, causes chemoresistance and contributes to epithelial mesenchymal transition IL6-Stat3 feed-forward loops amplify pro-tumorigenic and pro-metastatic signals in cancer cells 4,41. However, the role and importance of the IL6-Stat3 axis in s has not been studied before. We document that s can be actively involved in breast tumour metastasis as one of the orchestrators of metastasis via the IL6- Stat3 axis. We show that containing IL6 induces lymphatic expression of CCL5 in the pre-metastatic organs, forming chemotactic gradients to recruit CCR5-positive cancer cells into the organs (Figs 1 and 2). We measured the concentration of mccl5 in the plasma of mice treated with over a 2-week period (Supplementary Fig. 3). mccl5 increased with time of treatment, and the increasing trend was sustained for an additional week after stopping the treatment (maximum level ¼ 45 pg ml 1 ). Compared with the level of mccl5 (25 pg ml 1 ) in normal tumour xenograft models without treatment (Fig. 5d), pretreatment can create a dramatic CCL5 gradient in the system to facilitate tumour dissemination. -induced metastasis was blocked by maraviroc, a CCR5 inhibitor (Fig. 2). We also evaluated the therapeutic effects of maraviroc in spontaneous metastasis models without (Supplementary Fig. 9). Maraviroc treatment inhibited tumour metastasis, suggesting that the CCL5-CCR5 axis is also central in general and spontaneous metastasis models. The CCL5-CCR5 axis needs to be further investigated in murine tumour models as well, since our nude mouse models may have limitations with the absence of T lymphocytes that could be one of the mediators of metastasis. To establish the clinical relevance of our findings, we evaluated the IL6-CCL5 axis by analysing TCGA mrna-sequencing data sets (Supplementary Fig. 18). Both IL6 and CCL5 were significantly overexpressed in TNBC tumours over ER þ /PR þ / HER2 tumours (Supplementary Fig. 18a,b). This finding is consistent with our results that only TNBC cell lines (MB231 and SUM149) express IL6 and induce CCL5 expression in s and that the MCF7 cell line does not (Figs 1d and 6g). In addition, the expression of IL6 and CCL5 mrnas was significantly correlated in LN-positive TNBC samples over LN-negative samples, suggesting that the IL6 and CCL5 can serve as therapeutic and prognostic markers in TNBC metastasis (Supplementary Fig. 18c). The axis needs to be further studied in other subtypes of breast cancer and other cancers to expand the application. To expand on the discovery that organ-residing s promote metastasis via CCL5 expression (Figs 1 and 2), we examined whether s exist in the primary tumours, as well to orchestrate metastasis by actually connecting the primary tumours and distant organs. s were detected in the MB231 tumours; moreover, s within the tumour expressed CCL5 (Supplementary Fig. 21). This suggests that s can form a CCL5 gradient even in tumour stroma, which can trigger initial recruitment of cancer cells into the lymphatic system via intratumoural and peritumoural LVs. The presence of s in the tumour is due to tumour lymphangiogenesis, driven by tumour cell-secreted lymphangiogenic factors such as VEGF-C/ D 42,43. To expand on the classical understanding of tumour lymphangiogenesis, we add a new concept that s within tumours and distal organs can create chemotactic gradients to facilitate lymphogenous metastasis via the CCL5-CCR5 axis. We investigated mechanisms of CCL5 upregulation in s by. In previous studies of CCL5 regulation, tumour-necrosis factor-a, not IL6/gp13, induced CCL5 expression in vascular smooth muscle cells 44 in an NFkB-dependent manner; NFkBdependent CCL5 expression has also been studied in other types of cells In this study, however, we found that IL6-induced CCL5 is not colocalized with asma-positive cells (Supplementary Fig. 2b) and is not associated with an NFkB-Stat3 complex (Supplementary Fig. 16c,d). Instead, pstat3 forms a ternary complex consisting of pstat3, pc-jun and patf-2 in response to IL6, which regulates CCL5 expression in the lymphatic system; this mechanism has not been previously discovered. ChIP and EMSA experiments showed the binding of the ternary complex to the CRE site of the CCL5 promoter (Fig. 7d,e; Supplementary Fig. 19). We additionally tested the effect of EGF on CCL5 expression in s, as MB231-s express EGF (Fig. 3g) and EGF-derived Src pathways may contribute to activation of Stat3 pathways 29. EGF treatment phosphorylated c-jun and ATF-2, but not Stat3 in s, and did not induce CCL5 expression (Supplementary Fig. 22). Interestingly, unlike s, HUVECs could not be conditioned by (Fig. 3g; Supplementary Fig. 23). It has been reported that s express 3 4 times more gp13, compared with BEC 47. Gp13 is a co-receptor of IL6 receptor and plays a role as an IL6 signal transducer 48. We showed that the gp13-jak2 axis is a pivotal bridge for IL6-pStat3 signalling transduction in s (Supplementary Fig. 17), and consistently observed less gp13 as well as less pstat3 in BEC and no effect (Supplementary Fig. 23). Other molecular details of IL6-dependent induction of CCL5 remain to be elucidated. We showed that tumour-conditioned s also promote angiogenesis (Figs 3 and 4), which has not been reported before. While physiological s maintain angiogenic homeostasis, the -treated secretome is abnormal and highly angiogenic. Breast cancer involves metastasis to the LNs, thus the LNs need to 12 NATURE COMMUNICATIONS 5:4715 DOI: 1.138/ncomms & 214 Macmillan Publishers Limited. All rights reserved.